Logic, Reasoning and A Programming Language for Simulating Economic and Business Processes with Artificially Intelligent Agents

8 Logical Language and Informal Description

Logical language is much closer to a natural language description of agent behaviour than, say, Bayesian probability theory. Thus the declarative nature of SDML enables the specification of agent behaviour with the minimum of distortion. Similarly the results of the simulation are stored in the form of declared `facts' on the various databases, in the language of SDML and the clauses specifically designed by the modeller. The results are much easier to interpret, especially when they involve qualitative information. We do not claim that such interpretations are impossible with imperative and numerically based languages, but that they are more cumbersome to use in this way compared to the more natural manner afforded by a declarative framework and programming language.

Like any formal system, a large part of the benefit of the modelling process is that requires the underlying assumptions to be made explicit. Since the process of translation from an informal description or intuition to the language of SDML and back again in the interpretation of results is much easier and more natural in such a language, it is much more likely that both assumptions and the interpretation of the results will not be hidden by the formal machinery. We have found this approach to provide the basis for a more productive dialogue between modeller and model than is possible with imperative models and languages.

Similarly the results of an SDML model are transparently revealed in terms of true statements aiding interpretation of the results.

Such a style of modelling makes it easier to prove theorems about the limitations of model behaviour in each case -- a property which is important both for theoretical justification and practical understanding of the processes we are modelling (cf. Axtell and Epstein,1994). While SDML and the modelling process give insights into what is possible, providing the basis for new languages of discourse about the processes being investigated, theorem proving can enable statements to be made about what is necessarily true about of model.

We believe that this process of theorem proving can be substantially automated. Once the modeller has gained an understanding into the nature of the models, hypotheses can be made about that model and the automatic theorem prover can be used to attempt to prove them in the logic. This can also have the useful effect of revealing further assumptions necessary to prove the hypotheses*1.

Logic, Reasoning and A Programming Language for Simulating Economic and Business Processes with Artificially Intelligent Agents - 12 APR 96

[Next] [Previous] [Top] [Contents]

Generated with CERN WebMaker